[PATCH] D115726: [InstCombine] Fold for masked gather when loading the same value each time.
Caroline via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Fri Dec 17 06:18:40 PST 2021
CarolineConcatto updated this revision to Diff 395115.
CarolineConcatto marked 5 inline comments as done.
CarolineConcatto added a comment.
-address comments in the test
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D115726/new/
https://reviews.llvm.org/D115726
Files:
llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
llvm/test/Transforms/InstCombine/vscale_masked_intrinsics.ll
Index: llvm/test/Transforms/InstCombine/vscale_masked_intrinsics.ll
===================================================================
--- /dev/null
+++ llvm/test/Transforms/InstCombine/vscale_masked_intrinsics.ll
@@ -0,0 +1,49 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -instcombine -S < %s | FileCheck %s
+
+;; Splat Value and all active/one mask
+define <vscale x 2 x i64> @valid_invariant_load_i64(i64* %src) #0 {
+; CHECK-LABEL: @valid_invariant_load_i64(
+; CHECK-NEXT: [[LOAD_COMBINE:%.*]] = load i64, i64* [[SRC:%.*]], align 8
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[LOAD_COMBINE]], i32 0
+; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT1]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
+; CHECK-NEXT: ret <vscale x 2 x i64> [[BROADCAST_SPLAT2]]
+;
+ %broadcast.splatinsert = insertelement <vscale x 2 x i64*> poison, i64 *%src, i32 0
+ %broadcast.splat = shufflevector <vscale x 2 x i64*> %broadcast.splatinsert, <vscale x 2 x i64*> poison, <vscale x 2 x i32> zeroinitializer
+ %res = call <vscale x 2 x i64> @llvm.masked.gather.nxv2i64(<vscale x 2 x i64*> %broadcast.splat, i32 8, <vscale x 2 x i1> shufflevector (<vscale x 2 x i1> insertelement (<vscale x 2 x i1> poison, i1 true, i32 0), <vscale x 2 x i1> poison, <vscale x 2 x i32> zeroinitializer), <vscale x 2 x i64> undef)
+ ret <vscale x 2 x i64> %res
+}
+
+;; Not a splat value
+define <vscale x 2 x i64> @invalid_value_invariant_load_i64(i64* %src ) #0 {
+; CHECK-LABEL: @invalid_value_invariant_load_i64(
+; CHECK-NEXT: [[INSERT_VALUE:%.*]] = insertelement <vscale x 2 x i64*> poison, i64* [[SRC:%.*]], i32 1
+; CHECK-NEXT: [[RES:%.*]] = call <vscale x 2 x i64> @llvm.masked.gather.nxv2i64.nxv2p0i64(<vscale x 2 x i64*> [[INSERT_VALUE]], i32 8, <vscale x 2 x i1> shufflevector (<vscale x 2 x i1> insertelement (<vscale x 2 x i1> poison, i1 true, i32 0), <vscale x 2 x i1> poison, <vscale x 2 x i32> zeroinitializer), <vscale x 2 x i64> undef)
+; CHECK-NEXT: ret <vscale x 2 x i64> [[RES]]
+;
+ %insert.value = insertelement <vscale x 2 x i64*> poison, i64 *%src, i32 1
+ %res = call <vscale x 2 x i64> @llvm.masked.gather.nxv2i64(<vscale x 2 x i64*> %insert.value, i32 8, <vscale x 2 x i1> shufflevector (<vscale x 2 x i1> insertelement (<vscale x 2 x i1> poison, i1 true, i32 0), <vscale x 2 x i1> poison, <vscale x 2 x i32> zeroinitializer), <vscale x 2 x i64> undef)
+ ret <vscale x 2 x i64> %res
+}
+
+;; Not all active/one or inactive/zero mask
+define <vscale x 2 x i64> @invalid_mask_invariant_load_i64(i64* %src, <vscale x 2 x i1> %mask) #0 {
+; CHECK-LABEL: @invalid_mask_invariant_load_i64(
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64*> poison, i64* [[SRC:%.*]], i32 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64*> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64*> poison, <vscale x 2 x i32> zeroinitializer
+; CHECK-NEXT: [[RES:%.*]] = call <vscale x 2 x i64> @llvm.masked.gather.nxv2i64.nxv2p0i64(<vscale x 2 x i64*> [[BROADCAST_SPLAT]], i32 8, <vscale x 2 x i1> [[MASK:%.*]], <vscale x 2 x i64> undef)
+; CHECK-NEXT: ret <vscale x 2 x i64> [[RES]]
+;
+ %broadcast.splatinsert = insertelement <vscale x 2 x i64*> poison, i64 *%src, i32 0
+ %broadcast.splat = shufflevector <vscale x 2 x i64*> %broadcast.splatinsert, <vscale x 2 x i64*> poison, <vscale x 2 x i32> zeroinitializer
+ %res = call <vscale x 2 x i64> @llvm.masked.gather.nxv2i64(<vscale x 2 x i64*> %broadcast.splat, i32 8, <vscale x 2 x i1> %mask, <vscale x 2 x i64> undef)
+ ret <vscale x 2 x i64> %res
+}
+
+
+; Function Attrs:
+declare <vscale x 2 x i64> @llvm.masked.gather.nxv2i64(<vscale x 2 x i64*>, i32, <vscale x 2 x i1>, <vscale x 2 x i64>)
+
+attributes #0 = { "target-features"="+sve,+sve" }
+
Index: llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
===================================================================
--- llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -352,9 +352,27 @@
// * Dereferenceable address & few lanes -> scalarize speculative load/selects
// * Adjacent vector addresses -> masked.load
// * Narrow width by halfs excluding zero/undef lanes
-// * Vector splat address w/known mask -> scalar load
// * Vector incrementing address -> vector masked load
Instruction *InstCombinerImpl::simplifyMaskedGather(IntrinsicInst &II) {
+ auto *ConstMask = dyn_cast<Constant>(II.getArgOperand(2));
+ if (!ConstMask)
+ return nullptr;
+
+ // Vector splat address w/known mask -> scalar load
+ // Fold the gather to load the source vector first lane
+ // because it is reloading the same value each time
+ if (ConstMask->isAllOnesValue())
+ if (auto *SplatPtr = getSplatValue(II.getArgOperand(0))) {
+ auto *VecTy = cast<VectorType>(II.getType());
+ const Align Alignment =
+ cast<ConstantInt>(II.getArgOperand(1))->getAlignValue();
+ LoadInst *L = Builder.CreateAlignedLoad(VecTy->getElementType(), SplatPtr,
+ Alignment, "load.combine");
+ Value *Shuf =
+ Builder.CreateVectorSplat(VecTy->getElementCount(), L, "broadcast");
+ return replaceInstUsesWith(II, cast<Instruction>(Shuf));
+ }
+
return nullptr;
}
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D115726.395115.patch
Type: text/x-patch
Size: 5436 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20211217/e99c54b2/attachment.bin>
More information about the llvm-commits
mailing list